System Architecture

What MCP needs to work in production: context management, scalable infrastructure, and solutions to real-world scaling issues like those handled by Uber and Duolingo.

I discuss streams. It's a pretext to learn about higher-order streams, like `flatten` and to introduce a new stream: `switch`! It's very useful, and will have no secret for you.

Most systems don’t fail because of complex algorithms. They slow down because the same work is repeated too many times.A database receives thousands of identica...

Let’s explore Traefik’s internal architecture and see how it integrates with Kubernetes to monitor changes and update request routing dynamically. A simplified view of the main components involved in the routing of requests When the server starts, it initializes a set of entry points. Each entry point is a listening endpoint through which traffic enters Traefik. In simplified terms, an entry point contains: A listener that accepts incoming connections, A handler that processes requests and writes responses. The next important components are the configuration watcher and the providers. A provider monitors a specific configuration source, for example: Kubernetes or Docker, and feeds the configuration watcher. The configuration watcher maintains a set of callbacks and calls them when a new configuration arrives. One of the callbacks updates the current handler through the handler switcher.

Most teams use one signal to judge deploy risk: how big is this PR? It's intuitive. A 50-line change feels safer than a 2,000-line change. But it's also wrong — or at least deeply incomplete. Some of the highest-risk PRs ever merged into production...

Back on April 4, the social media site Bluesky suffered a pretty big outage. I was delighted to discover that one of their engineers, Jim Calabro, published a public writeup about it: April 2026 Ou…

Event-Driven by Oskar Dudycz

Why you need nested, structured error data.

An AI agent implements a feature. The code compiles. The tests pass. It still misses the point. The wrong kind of correct. Most of our software tooling is optim

At Slack, we recently unified fragmented probing into a standardized system that improves reliability, simplifies operations, and enables better measurement of network performance at scale.

Aegis is a fully open-source FPGA, from the silicon up. Open-source FPGA efforts have made huge strides: projects like Project IceStorm and Apicula reverse-engineer proprietary bitstream formats, O…

For some types of embedded systems — especially those that are safety-critical — it’s considered bad form to dynamically allocate memory during operation. While you can usually ar…

Kiki bills itself as the “array programming system of unknown origin.” We thought it reminded us of APL which, all by itself, isn’t a bad thing. The announcement post is decidedly…

You probably don’t think about it much, but your PC probably has a TPM or Trusted Platform Module. Windows 11 requires one, and most often, it stores keys to validate your boot process. Most …

KV caches are one of the most critical techniques for efficient inference in LLMs in production.

Large language models (LLMs) aren’t actually giant computer brains. Instead, they are massive vector spaces in which the probabilities of tokens occurring in a specific order is encoded. Bill…

A funny thing happened on the way to writing our book Architecture as Code—the entire industry shifted. Generally, we write books iteratively—starting with a

I spend a lot of time looking at alerts that say "possible MITM activity" or "ARP anomaly detected." They show up in the SIEM, get triaged, and usually turn out to be benign — a NIC failover, a miscon

webSocketSession.send("call someone") sends raw bytes over a persistent connection. That's it. No routing, no subscriptions, no message…

How coding agents use tools, memory, and repo context to make LLMs work better in practice

Journey from classic BPF in the nineties to the modern 64-bit virtual machine.

Extract docs from BEAM bytecode, embed with Ortex, search with sqlite_vec, serve through MCP to Claude Code. Everything local.

Railway is a full-stack cloud for deploying web apps, servers, databases, and more with automatic scaling, monitoring, and security.

Caution

CertKit automates your entire certificate lifecycle. Issue certificates via ACME, deploy them with the CertKit Agent, and verify everything with real TLS checks. No open ports, no ACME on your servers, no DNS changes.

Everything your CRDT paper didn't tell you about actually building one

Easy and intuitive Kalman Filter tutorial